I'm working on a meta-analysis, and I have a special problem regarding calculating effect size. Can anyone lend me some insight?

I have a treatment and a control group, and for each group I have the number of adverse events (not the number of people in each group experiencing the event, but the raw count of events). A single patient may experience multiple instances of the adverse event. Incidence rather than prevalence. I'm trying to figure out how to calculate effect size with this. I can't use odds-ratio because the count of adverse events is continuous and not dichotomous, and I can't use mean difference because the number of adverse events is not tracked at the level of the individual subject. Any ideas? Thanks!


Suppose you have n people in each group and they experience a total of r events. You can then calculate the average number of events per person (r/n). Now suppose you are willing to make an inferential leap and say that the number of events per person is distributed as Poisson. You now know the Poisson parameter (usually denoted $$\lambda = r/n$$ lambda) so you can estimate the number with zero events, and hence the proportion who experienced at least one, possibly more events.

  • $\begingroup$ Many thanks! We'll see if the Poisson assumptions satisfy the review board. $\endgroup$
    – wag
    Mar 14 '16 at 22:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.